RAG vs OKF (Open Knowledge Format) logopeterindia.net
Knowledge Hub

RAG vs OKF (Open Knowledge Format)

Retrieval at scale versus curated markdown knowledge: what Retrieval-Augmented Generation and the Open Knowledge Format each do, and how agents can use both.

46 topics•7 categories•Updated October 2026

RAG and OKF: Two Layers of Agent Knowledge

2 entries

Retrieval-augmented generation (RAG) searches a large body of documents at question time: content is split into chunks, embedded, indexed, and the most similar chunks are passed to the model. The Open Knowledge Format (OKF) is a different idea. Google Cloud published it in June 2026 as a vendor-neutral format for representing knowledge as plain markdown files with YAML frontmatter, linked together so an agent can read the exact concept it needs. In short, RAG is built for finding information in a large, changing corpus, while OKF is built for stating curated knowledge, such as definitions, policies and runbooks, in a form that can be read directly, reviewed and versioned. They are complementary rather than rivals: curated, stable facts suit OKF bundles, and the large, fast-changing long tail stays in a retrieval pipeline. An OKF bundle can itself be indexed by a RAG system. The framing of this page follows an October 2026 article by Rohan Mistry in Towards AI; the text here is written independently, and specifics should be checked against the specification.

01Source articleRAG vs OKF: Stop Choosing One. Why AI Agents Need Both, by Rohan Mistry, Towards AI (member-only story). 02RAG paper (Lewis et al.)The 2020 paper that introduced retrieval-augmented generation.

Open Knowledge Format (OKF)

9 entries

The specification, tooling and explainers.

03OKF repositoryOfficial Google Cloud repository with the specification, a proof-of-concept agent that produces bundles, sample bundles and a graph visualizer. 04OKF specification (SPEC.md)The authoritative definition of bundles, concept files and frontmatter. Explainers disagree on details, so check this document. 05Google Cloud Tech announcementGoogle Cloud's post introducing OKF as a portable format for the LLM-wiki pattern. 06GitBook: What is OKF?Explainer on OKF and how documentation platforms relate to it. 07Wavect: OKF enterprise guideThird-party enterprise guide to OKF. 08innFactory: OKF standardThird-party explainer on OKF as a standard for AI knowledge. 09WitsCode: OKF guideThird-party complete guide to OKF. 10Grounding Page: OKFThird-party fact sheet on OKF. 11MarkTechPost reportNews report on the release; the site blocks automated checks.

The LLM Wiki Idea

5 entries

The pattern OKF formalizes.

12Karpathy's LLM Wiki gistAndrej Karpathy's idea file on having an LLM build and maintain a persistent, interlinked markdown wiki instead of rediscovering knowledge on every question. 13ObsidianMarkdown note-taking app often used to browse and edit wiki-style knowledge bases. 14Markdown guideReference for markdown syntax. 15YAMLThe data format used for frontmatter. 16GitVersion control used to track changes to knowledge files.

RAG Foundations & Frameworks

8 entries

Build and evaluate retrieval pipelines.

17LlamaIndexFramework for building RAG and agent applications over your data. 18LangChain RAG tutorialTutorial for building a RAG application. 19HaystackOpen-source framework for search and RAG pipelines. 20Anthropic: Contextual RetrievalAnthropic's write-up on improving retrieval by adding context to chunks. 21RAG TechniquesCommunity collection of RAG techniques and notebooks. 22RagasEvaluation framework for RAG applications. 23Ragas repositorySource code for Ragas. 24Hugging Face: RAG evaluationCookbook recipe for evaluating RAG.

Vector Databases & Search

6 entries

Where RAG indexes live.

25pgvectorVector similarity search for PostgreSQL. 26QdrantOpen-source vector database. 27MilvusOpen-source vector database. 28WeaviateVector database with hybrid search. 29ChromaOpen-source embedding database. 30PineconeManaged vector database.

Knowledge Graphs & Agent Memory

8 entries

Other ways to give agents structured or persistent knowledge.

31Microsoft GraphRAGGraph-based approach to RAG from Microsoft Research. 32GraphRAG repositorySource code for GraphRAG. 33LightRAGGraph-enhanced RAG framework. 34Neo4jGraph database often used for knowledge graphs. 35GraphitiFramework for temporal knowledge graphs that serve agent memory. 36Mem0Memory layer for AI agents. 37LettaFramework for stateful agents with memory. 38Model Context ProtocolOpen protocol for connecting agents to tools and data, including file and knowledge servers.

Related Pages

8 entries

Neighbouring pages on this site.

39RAG Architectures ElucidatedRAG types, architecture patterns and chunking strategies. 40Vector DatabasesVector database directory. 41Vector Databases for Enterprise AIEnterprise vector databases. 42Knowledge GraphsKnowledge graph tools and resources. 43Knowledge EngineeringKnowledge engineering. 44Top MCP ServersMCP servers directory. 45Agentic AIAgentic AI hub. 46Generative AIGenerative AI hub.

Frequently Asked Questions

What is OKF?

The Open Knowledge Format is a vendor-neutral specification from Google Cloud for representing knowledge as a directory of markdown files with YAML frontmatter. Each file describes a concept, and links between files let an agent navigate related concepts. See the specification in the official repository for the exact rules.

How is OKF different from RAG?

RAG searches a large corpus by similarity and returns likely-relevant chunks. OKF stores curated knowledge that an agent reads directly by following links from an index. RAG favors scale and discovery; OKF favors precision, explicit relationships and reviewability.

Does OKF replace RAG?

No. Curated, stable knowledge such as definitions, policies and runbooks suits OKF, while large, fast-changing or unstructured collections still need retrieval. Many designs combine them, for example by routing questions about known facts to the curated bundle and everything else to search.

What is the LLM Wiki?

A pattern described by Andrej Karpathy in which an LLM compiles source material into a persistent, interlinked markdown wiki and keeps it updated, rather than retrieving from raw documents on every question. OKF formalizes this style into a portable format.

Do I need special tools for OKF?

The format is plain markdown with YAML frontmatter, so a text editor and a version-controlled folder are enough to start. Parsers, validators and viewers exist, but check each project's maturity before relying on it.

How current is this page?

Links were checked in October 2026. OKF is a young specification and several explainers disagree on details such as required fields and version dates, so use the specification repository as the authority. Sites that block automated checks are kept and may need a manual click.